# Windows 11 — TensorFlow GPU Installation

### Links

* Official TF installation guide:
[https://www.tensorflow.org/install/pip](https://www.tensorflow.org/install/pip)
* Video tutorial:
[https://www.youtube.com/watch?v=eErbMsP8v9I](https://www.youtube.com/watch?v=eErbMsP8v9I)

* protobuf compatibility
https://stackoverflow.com/questions/76641641/is-there-any-reference-to-tensorflow-datasets-compatibility-chart-specifically

---

## 1. Clean up old environment(optional)

conda deactivate
conda env remove -n tf -y
conda env list

---

## 2. Create a fresh Conda environment

conda create -n tf python=3.9 -y
conda activate tf

---

## 3. Install CUDA + cuDNN (Conda versions compatible with TF 2.10)

conda install -c conda-forge cudatoolkit=11.2 cudnn=8.1.0 -y

---

## 4. Install required Python packages

pip install notebook
pip install tensorflow==2.10.1
pip install numpy==1.26.4
pip install "protobuf==3.20" --force-reinstall
pip install "grpcio<=1.48" --force-reinstall
pip install tensorflow-addons==0.18.0
pip install numpy==1.26.4

---

## 5. Remember to **Reload Window** in VS Code

After installing packages, reload VS Code to ensure that Jupyter kernels and Python executables correctly detect the newly installed packages.


---

## 6. About the *protobuf* version conflict

I frequently encountered the protobuf installation warning:

* TensorFlow 2.10 requires **protobuf < 3.20**
* But *tensorflow_datasets* usually prefers a newer protobuf version
* I found that installing **protobuf==3.20** surprisingly works fine for both packages

It was a strange conflict, but installing exactly `3.20.0` resolved the issue.


---

## 7. Test your installation

Use this script to verify all package versions and GPU availability:

import tensorflow as tf
import tensorflow_addons as tfa
import tensorflow_datasets as tfds
import numpy as np
import importlib.metadata as importlib_metadata  

def get_version(pkg_name: str) -> str:
    """Safely return package version, or '(not installed)' if missing."""
    try:
        return importlib_metadata.version(pkg_name)
    except importlib_metadata.PackageNotFoundError:
        return "(not installed)"


print("TF version:", tf.__version__)
print("TFA version:", tfa.__version__)
print("TFDS version:", tfds.__version__)
print("NumPy version:", np.__version__)
print("protobuf:", get_version("protobuf"))
print("GPU devices:", tf.config.list_physical_devices("GPU"))

### ✔️ Expected output example:

TF version: 2.10.1
TFA version: 0.18.0
TFDS version: 4.9.2
NumPy version: 1.26.4
protobuf: 3.20.0
GPU devices: [PhysicalDevice(name='/physical_device:GPU:0', device_type='GPU')]

### modify conda activation script
C:/ProgramData/anaconda3/Scripts/activate